Development of a prognostic model for breast cancer patients based on intratumoral tumor-infiltrating lymphocytes using machine learning algorithms.

BACKGROUND

Breast cancer remains a formidable global health challenge, with tumor-infiltrating lymphocytes (TILs) serving as pivotal biomarkers associated with disease progression, therapeutic response, and survival. While research typically focused on stromal TILs (sTILs), we hypothesize that intratumoral TILs (iTILs), which are in direct contact with tumor cells, have a more profound role in the immune-tumor interactions. In light of this, we have developed an iTIL-centric model for breast cancer patient stratification and prognostic prediction.

METHODS

We sourced RNA-seq data and clinical profiles of breast cancer patients from The Cancer Genome Atlas (TCGA) and the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) to form our training dataset. Testing datasets, including GSE20685, GSE42568, GSE48390, and GSE88770, were retrieved from Gene Expression Omnibus (GEO). Employing consensus clustering and Weighted Correlation Network Analysis (WGCNA), we identified iTIL-associated hub genes. Our iTIL-centric signature was developed using a machine learning framework integrating 101 algorithms, validated across independent testing sets. Kaplan-Meier analysis and a nomogram model were utilized to evaluate the prognostic accuracy and clinical correlation of our model. GO and KEGG analyses elucidated the biological processes and pathways related to the iTIL signature. The immune profiling provided a comprehensive assessment of the immunological landscape. Moreover, potential drugs for high-risk patients were identified using CTRP v.2.0 and PRISM databases.

RESULTS

Our study constructed a pioneering prognostic model based on iTIL-centric signature via a machine learning framework that evaluated 101 algorithm combinations. This model revealed significant differences in the immune landscape among stratified patient cohorts, and demonstrated robust predictive capabilities across multiple datasets. The model showed excellent predictive performance with area under the curve (AUC) values of 0.940, 0.959, and 0.973 for 3-, 5-, and 10-year survival predictions, respectively. Additionally, it was identified as a significant risk factor for overall survival (OS) in the univariate analysis, with a hazard ratio (HR) > 1 and a p-value < 0.001.

CONCLUSIONS

Our prognostic model, founded on machining learning algorithms and anchored by an iTIL-centric signature, stands out as an invaluable tool for breast cancer patients, offering advanced prognostic insights and facilitating the development of personalized therapeutic strategies.